◀ Back to the quest log주석야금
CLEARQUEST 01HR tech · hiring-match platform

Tearing down the filter wall with a conversational agent

Redesigned a hiring-match search that used to demand 4–5 conditions on one screen into a step-by-step conversational flow where the AI asks back and shows its reasoning.

Industry
HR tech · hiring-match platform
Scope
UX audit · flow redesign · wireframes
Timeline
Jan 31 – Feb 6, 2026 (1 week)
Redesigned wireframe · searching the candidate pool after the questions
Redesigned wireframe · searching the candidate pool after the questions

Loot

Entry questions3 turnsBefore: 4–5 conditions entered at once on one screen
Audit3 × 33 as-is problems · 3 to-be strategies
Timeline1 weekFrom audit to build-ready wireframes

The request

Recruiters had to type 4–5 conditions — domain, role, skills, education — into the search screen all at once. Filter chips filled the screen and pushed the AI recommendations out of view, and long prose summaries made it hard to see why a candidate matched.

How we solved it

We turned it into a conversation where the AI asks for conditions in three short turns. Candidate cards got fit badges and keyword highlights, so the reason for each match is visible at a glance.

Before — one screen that asks everything

  1. 01Enter 4–5 conditions at once
  2. 02Filter chips fill the screen
  3. 03AI picks pushed below the fold
  4. 04Dig through long prose for reasons

After — an agent that asks step by step

  1. 01Core criteria
  2. 02Specific skills
  3. 03Extra conditions
  4. 04Fit badges & keyword highlights
  5. 05Message straight from the list
LOG

The problem

On the search entry screen, recruiters were asked to specify every candidate condition (domain, role, skills, education) as free text all at once. Setting filters manually or uploading a job description filled the screen with filter chips, pushing the AI’s recommendations and conversational context further down. Result summaries were also long blocks of prose, making it hard to see at a glance how the entered conditions actually shaped the results.

Approach

“Agent UX over tool UX” — the system’s underlying logic stayed the same; the screens got radically simpler.

  • Entry flow: instead of asking everything at once, redesigned it as a 3-turn conversation — core conditions → detailed skills → secondary conditions.
  • Trust visualization: a fit badge at the top of each candidate card, with keywords mentioned in conversation highlighted so matching evidence is visible at a glance — full reasoning stays behind a popover on icon click, keeping the default view simple.
  • Real-time feedback: an “Analyzing…” message with a skeleton UI conveys the AI working in real time, and messages can be sent straight from the list without opening the detail page, cutting click depth.

Diagrammed the existing and improved flows side by side to compare what changed and why, and overlaid both states (as-is/to-be) on real UI screenshots to diagnose item by item.

Delivered

  • UX diagnosis report (3 as-is problems, 3 to-be strategies)
  • Improved user flow diagrams (before/after comparison)
  • Development-ready wireframes with interaction specs per screen element